Numerous test data accumulates in the process of gas reservoir exploration and development, so it is necessary to apply the data mining technology to this process. Influenced by the geologic factors such as structure, deposition and diagenesis, tight sandstone gas reservoir formation types are so diversified that traditional cross-plot analysis technique hardly identify the formation types. In this paper, the formation types of tight sandstone gas reservoir in Daniudi area are successfully identified using the decision tree algorithm of data mining based on hierarchical decomposition theory, facilitating the development of the gas reservoir.

Abstract. Domestic oil-gas fields are almost approaching production tail, and an increasing number of non-traditional oil-gas reservoirs are encountered during the process of exploratory development, which leads to a urgent requirement for an advanced method in that conventional methods, such as cross plot and multiple linear regression cannot precisely describe such complex oil-gas reservoirs. Thus, the main purpose of this paper is to come up with method of Decision Tree as final model for identification of reservoir fluid based on the comparison of advantage and disadvantage of fours methods, including Decision Tree, Support Vector Machines, Artificial Neural Network and Bayesian Network. In sum, data mining is a prospective applied method in oil reservoir geology.